Traditional clinical monitoring of children with epilepsy does not appear to be sufficiently sensitive to cognitive functioning and behavioral problems. Although subtle, these changes may alter a child's ability to perform well in school and in society. Physicians must prevent seizures without producing intolerable side effects, and ways of more appropriately assessing these side effects must be developed. In this double-blind, counter-balanced, crossover study of 21 children, the effects of phenobarbital and valproic acid on cognitive functioning and behavior were measured. There was no difference in seizure control between the drugs, and each medication was maintained in the therapeutic range for 6 months (mean phenobarbital level, 21.2 micrograms/mL; mean valproic acid level, 94.1 micrograms/mL). Children were treated with each drug for 6 months. Differences between the drugs were seen on measurements of cognitive function and behavior. On four tests of neuropsychologic function, children performed significantly less well while receiving phenobarbital (P less than .01). There was no evidence that the patients were sedated or less able to perform continuous performance tasks while receiving phenobarbital. Parental assessment of behavior indicated significantly worse behavior with the phenobarbital regimen for three items (P less than .01) and children were measurably more "hyperactive" (P less than .05). Routine clinical assessment of the patients did not reveal differences between the drugs with respect to routine laboratory measurements or side effects as assessed by history or physical examination. Although children may appear to tolerate a medication without clinically apparent problems, subtle but significant changes in intellectual function and behavior may be occurring. Additional, more sensitive, methods of monitoring patients while receiving these drugs is necessary.